Reference // Glossary
The terms we use, in plain language.
A short, honest glossary of the engineering, AI, and data terms that come up in our work — defined the way we actually mean them.
- RAG (Retrieval-Augmented Generation)
- A technique where a language model retrieves relevant documents from a trusted data source at query time and uses them to ground its answer, instead of relying only on what it memorized during training. RAG is how production AI agents cite sources and stay accurate against private data.
- LLM Agent
- Software that uses a large language model to complete multi-step tasks: reading context, calling tools and APIs, and acting within defined guardrails. Unlike a chatbot, an agent owns an outcome — resolving a ticket, reconciling a record — not just a conversation.
- Eval Harness
- A regression test suite for AI behavior. It scores an agent's outputs against known-correct examples so that changes ship when the evals pass, not when a demo feels good. Without one, you can't tell whether an AI system is improving or quietly degrading.
- Vector Database
- A datastore that indexes text and other data as numeric embeddings so it can be searched by meaning rather than exact keywords. Vector databases like pgvector, Pinecone, and OpenSearch power the retrieval step in RAG systems.
- Lakehouse
- A data architecture that combines the low-cost, open storage of a data lake with the structure and transactional guarantees of a warehouse, usually via table formats like Apache Iceberg or Delta Lake. It lets analytics and machine learning run on the same governed data.
- Data Warehouse
- A system optimized for analytical queries over large volumes of structured, modeled data — for example Snowflake or BigQuery. A warehouse is where dashboards, reports, and business metrics are computed.
- Change Data Capture (CDC)
- A technique that streams every insert, update, and delete from a source database to downstream systems in near real time, instead of re-copying whole tables on a schedule. CDC keeps pipelines and warehouses fresh without heavy batch jobs.
- Strangler Pattern
- An incremental migration strategy where a new system is built around the edges of a legacy one and gradually takes over its responsibilities, route by route, until the old system can be retired. It avoids risky big-bang rewrites and keeps the product shipping throughout.
- Feature Store
- Infrastructure that computes, stores, and serves the input features a machine-learning model needs, consistently between training and live serving. It prevents the train/serve skew that silently degrades model quality.
- Semantic Layer
- A modeling layer that defines business metrics — revenue, active users, churn — once, in one place, so every dashboard and query computes them the same way. It is what stops two reports from disagreeing about the same number.
- Observability
- The practice of instrumenting a system with logs, metrics, and traces so its internal state can be understood from the outside. It is essential for knowing whether software is healthy, degrading, or failing in production.
- RBAC (Role-Based Access Control)
- An access model where permissions are attached to roles and users are granted roles, rather than permissions being assigned to each user individually. It makes who-can-do-what auditable and is foundational to multi-tenant and internal tools.
- Fractional CTO
- A senior engineering leader engaged part-time to own architecture decisions, hiring, and technical strategy for a company that needs that judgment before it can justify a full-time executive hire.
- WCAG 2.1 AA
- The Web Content Accessibility Guidelines, level AA — the conformance target most U.S. public-sector and enterprise software is expected to meet so the product is usable by people with disabilities. Kevadia builds toward it from the first sprint, not as a retrofit.
- Small Business Enterprise (SBE)
- A state certification identifying a qualifying small business, which can make it eligible for small-business set-asides and preferences in government procurement. Kevadia is certified as an SBE by the State of New Jersey.
- Minority Business Enterprise (MBE)
- A certification identifying a business owned and controlled by members of a recognized minority group, used in supplier-diversity and set-aside programs. Kevadia is certified as an MBE by the State of New Jersey.
- Technical Due Diligence
- An independent assessment of a company's architecture, code quality, team, scaling risk, and technology claims — typically commissioned by an investor or acquirer before a deal. The output is a written report in business language: what's real, what's fragile, and what it costs to fix.
- Legacy Modernization
- The process of replacing or refactoring aging software — for example COBOL or VB6 systems — into maintainable, well-tested modern services, usually incrementally so the system keeps running throughout.